James writes:

> Re: TCP/IP & IP "Precedence" bit.

Ah HA!

> I have run into an interesting case where the 3000 sends a RESET and
> clears the TCP connection when a client increases the precedence in an
> IP packet sent to the 3000 NS-TRANSPORT.

I have a customer who complains of groups of inbound NS-VT users all
getting dumped by the 3000 simultaneously (typically five users will
have there communications terminated over a few seconds).  Exploring the
NMLG files we find "Precedence Fault" errors followed by a VT disconnect.

So far I have been unable to find any document or person at HP who could
explain this, other than the obvious connection to the IP header flag's
"precedence bits".

In this case the clients are all using various versions of Reflection.  At
one point the user I was talking to had his own (idle) session disconnect
while I was on the phone with him.

I believe it is likely that all of the users are going through some
reasonably new Cisco routers, and my last suspicion was that there was
some case where the Cisco would [expletive deleted] with the precedence
bits in the IP headers for a whole group of packets for some reason.

The customer reported that the problem would occur between 0 and 3 times
per day, each time a group of VT users would be dumped simultaneously for
no reason that they could detect.

> I have always believed it was illegal (against TCP/IP RFC specification)
> to change the precedence in an IP packet once the precedence has been
> negotiated in the 3-way TCP connection establishment SYN -> SYN/ACK -> ACK.

Mee Tooo.

> I back this up with RFC 793 which say's:

I like the other RFC (forget which one) which implies that the precedence
bits themselves are reserved for the Department of Defense and dire things
will happen to you if you try to do anything with them :-)

>   I would appreciate any input from my esteemed colleagues out here on the
> 3000-L news group, especially if you can find a document which overrides
> or changes my understanding of the above RFC 793.

Personally I think an interrogation of Cisco might be revealing.  Any of our
Cisco Certified Gurus on the list know anything about this?

G.